What it does
Story Keeper helps families keep what an elder said, not a summary of it. You point it at an interview recording on the NAS. It finds the file, transcribes it, saves the transcript, drafts a story outline, and checks every quoted passage word for word against the transcript. A quote it cannot find is blanked and marked "quote not verified".
The result is a saved text story index: one row per story with a title, a start time where the transcription provides one, the people named, and one exact quote, followed by an "Ask next time" list of questions the interview left open.
If the user wants, it also lists the people already named in the Photos app, proposes which named person matches each name in the stories, and waits for the user to confirm every pair before finding up to three photos per person.
How it is built
A Markdown-only orchestration Skill that delegates one step at a time to the file, knowledge, document and photo experts, and passes downstream only real completed results. It saves only after approval, never deletes, moves or renames anything, and never names faces.
Status and challenges
Drafted and statically checked with the supplied kit (zero errors, zero warnings), not yet run on a NAS. The package lists what the finalist test day must confirm: whether transcript segments carry times and speaker labels, whether the document expert can read the saved text file, which audio formats and languages are accepted, and whether the full transcript reaches the quote check. The quote check is a model comparison, so it is described as "checked", never "guaranteed".
What is next
Test on a real UGREEN AI NAS if selected as a finalist, using the test plan in the package: a ten-minute two-voice interview, a run with two named people in Photos, and an edge case where nobody is named in Photos.
Built With
- claude
- markdown
- ugos
- uliya
- uliya-skill-kit
